Zero2UI

How I fixed my transcript workflow in 26 minutes

Turning my videos into transcripts was a mess, and I had no time to learn the proper way.
So I did the lazy designer thing. I described the problem to Claude Fable 5 in plain words and let it build the tool. No code from me. The thing works, it solves a real problem in my week, and it cost something real, not nothing.

If you're learning design and the tech side at the same time, the channel is about going from zero to a shipped product. Subscribe if that's your path.

Figma agent: I scored it X out of 10

Figma agents just landed, so I am stress-testing them one by one. First test: I asked the agent to fill a whole screen with realistic content for a food delivery app and keep the layout and spacing exactly as they were.

7 out of 10. The spacing held up properly. The imagery was weak (do that part in another AI tool). And it still will not name your layers for you, that one is on you.

Grab the example file from our Figma community and run the same prompt yourself.

Video 1 of 10. What would you score it? Tell me below.

Anthropic just released Fable 5, their most powerful AI model yet. Here's what it actually means for designers.

It's better at coding, vision, and turning an idea into a working product than anything they've shipped before. But the thing it still can't do is taste. AI can do perfect. It can't do imperfect on purpose, with a point of view. That part is still yours.

Test it, form your own opinion, don't panic about the headlines.

Save this if you're a designer working out where AI tools actually fit. Full breakdown on the 02UI Youtube channel SOON.

Figma just shipped the thing every design system person has been asking for. And of course I can’t use it.

Check designs scans your file and catches every hard coded colour, every random spacing value, every corner radius that should have been a variable. Then it suggests the right one from your library and you apply it in one click. No more digging through 400 layers to find the one that quietly broke your system.

Here’s my problem. It’s Organization and Enterprise only, locked in a private beta. I pay for 3 full seats and I still can’t open it. So I get to make a video about a feature I can only watch other people use. Brilliant :)

If you maintain a design system, this is the cleanup tool you actually want. Just check your plan before you get your hopes up (most of us are not on the list).

Save this for when it reaches the rest of us.

My landlord’s story and my Figma file, same lesson…

656 gaps and paddings to fix in one Figma file. By hand that’s most of a day gone, and a boring day at that.

This was part 2 of a design system audit for a client. 2 years ago you tokenised all of that by hand because there was no other way to keep a system consistent as it grew. Today nobody wants to do it like that, and nobody has to.

So I recorded a voice note explaining the context, dropped it into Claude Code, turned it into a project file, connected to Figma through the API and let it run. 15 minutes. 0 mistakes (I spot-checked them, not all 656, but enough random ones to trust it).

I’m tool agnostic. I don’t marry any of them. But I optimise my workflow every single day, and this is what that looks like now.

Have you had one of these moments yet? The thing you used to grind through that suddenly takes minutes. Tell me below.

How to audit a design system in 6 steps.

Tokens are only the first of 6 dimensions in a real design system audit. Here's the full order I use, and where Claude Code actually earns its place.

The 6 dimensions, in order:

1- Tokens and variables (primitive vs semantic and brand)
2- Component architecture (variants, props, detached instances)
3- Accessibility (contrast, focus indicators, target sizes, state visibility)
4- Design and code parity (align with your developer)
5- Documentation (when to use each token and component)
6- Governance and publishing (versioning, adoption, ownership)

The prompt I used to start the audit is linked below. Copy it and run it in your favourite AI tool.

5 habits that fixed my Claude usage limit

Claude re-reads your entire conversation on every single turn. By message 30, one question can cost 50,000+ tokens. That's why you're hitting the wall. Here are 5 fixes I use daily.

My site scored 33 on an AI readiness audit. Here is the fix.

AI tools are already reading your website. Whether they understand what they find is a different question.

Google published official guidance this week confirming traditional SEO still works for their AI search features. But direct AI agents like Claude and Cursor work differently, and most websites have never prepared for them. This video explains the difference, shows you a free audit tool, and documents every fix I made to my own site after scoring 33 out of 100.

Full breakdown and all resources at 02ui.com/aeo